NeuroDETR: Multiscale Discrete Spatiotemporal Tokenization for Asynchronous EEG Event Detection

Yaozhi Wen; Kun Zhao; Wei Wei; Haoyang Chen; Kangning Wang; Yi Sun; Shaowei Gu; Dingkun Liu

cs.LGOct 14, 2026

NeuroDETR: Multiscale Discrete Spatiotemporal Tokenization for Asynchronous EEG Event Detection

Authors: Yaozhi Wen, Kun Zhao, Wei Wei, Haoyang Chen, Kangning Wang, Yi Sun, Shaowei Gu, Dingkun Liu,

Organizations: Laboratory of Brain Atlas and Brain-Inspired Intelligence, State Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology, Institute of Automation, Chinese Academy of Sciences · School of Future Technology, University of Chinese Academy of Sciences · Zhongguancun Academy · School of Artificial Intelligence, University of Chinese Academy of Sciences ·

Abstract

Asynchronous EEG event detection supports practical brain-computer interfaces by identifying event classes and temporal boundaries directly from continuous recordings without predefined event boundaries at inference time. The primary challenge lies in extracting event-related patterns embedded in ongoing background activity. To address this, we propose NeuroDETR, a discrete spatiotemporal multiscale framework that combines structured tokenization with event-aware feature refinement. Multiscale temporal codebooks capture brief responses and sustained dynamics, while Spatial-Pyramid VQ hierarchically encodes spatial structure across EEG channels. These complementary representations form a compact event memory for jointly classifying events and estimating their onsets and offsets. Before tokenization, we apply FOCAL, an event-aware sparse attention module that uses predicted event presence to guide temporal feature refinement and spatial aggregation, emphasizing likely event periods while preserving context across the recording. Experiments on eight EEG datasets that span diverse paradigms demonstrate state-of-the-art performance. These findings provide a basis for the development of BCI systems that recognize meaningful neural activity during continuous interaction and monitoring.

Figures & tables

Panel of Figure 1 — Figure 1: EEG decoding paradigms and relative performance. (a) Sliding-window classification and query-based event detection. (b) Performance of NeuroDETR and six baselines across eight datasets. Scores on each dataset are normalized by NeuroDETR’s score (NeuroDETR = 1).

Figure 2: Overview of NeuroDETR. (a) Continuous EEG input. (b) FOCAL uses event-presence estimates to guide local attention. (c) Temporal features are quantized at multiple resolutions. (d) Spatial-Pyramid VQ quantizes EEG features at global, regional, and local spatial levels. (e) A query decoder predicts event classes and boundaries from fused spatiotemporal memory. Detection and auxiliary losses provide training supervision.

	Method	P300		Motor Imagery		Motor Execution		Speech	Seizure
	Method	BI2014b	BI2013a	BNCI2014	BNCI2015	BNCI2020	HDOP	Broderick	TUSZ
F1 Sample	Random	$35.46_{\pm 0.60}$	$40.40_{\pm 0.36}$	$20.06_{\pm 0.54}$	$33.75_{\pm 1.12}$	$33.50_{\pm 0.50}$	$25.11_{\pm 0.39}$	$48.94_{\pm 0.15}$	$19.67$
	DyFADet	$34.84_{\pm 9.46}$	$\underline{55.08}_{\pm 2.22}$	$34.58_{\pm 20.67}$	$63.68_{\pm 3.52}$	$45.99_{\pm 22.05}$	$51.40_{\pm 1.23}$	$47.78_{\pm 2.66}$	$25.60$
	ActionFormer	$52.18_{\pm 1.65}$	$54.71_{\pm 2.29}$	$42.38_{\pm 4.85}$	$53.04_{\pm 1.29}$	$53.50_{\pm 2.16}$	$46.63_{\pm 1.80}$	$42.75_{\pm 1.17}$	$22.56$
	EventNet	$44.04_{\pm 2.74}$	$51.61_{\pm 5.88}$	$\underline{53.64}_{\pm 3.50}$	$\underline{73.94}_{\pm 4.02}$	$\textbf{59.91}_{\pm 2.94}$	$57.52_{\pm 3.82}$	$\underline{52.18}_{\pm 0.61}$	$24.52$
	DOSED	$38.56_{\pm 2.30}$	$44.75_{\pm 3.01}$	$46.06_{\pm 2.98}$	$60.85_{\pm 3.55}$	$55.36_{\pm 2.93}$	$\underline{64.15}_{\pm 1.69}$	$43.30_{\pm 0.44}$	$18.27$

Table 1: Results on eight EEG datasets: mean

\pm

SD over five cross-validation folds, except TUSZ, which uses a single result on its predefined split. Best and second-best scores are bold and underlined , respectively.

Variant	P300		Motor Imagery		Motor Execution		Speech	Seizure
Variant	BI2014b	BI2013a	BNCI2014	BNCI2015	BNCI2020	HDOP	Broderick	TUSZ
M1	$43.03_{\pm 1.75}$	$52.77_{\pm 1.22}$	$43.11_{\pm 6.70}$	$65.17_{\pm 5.33}$	$39.49_{\pm 3.13}$	$48.12_{\pm 3.84}$	$38.02_{\pm 0.43}$	$\underline{21.32}$
M2	$43.88_{\pm 2.46}$	$53.46_{\pm 0.99}$	$43.38_{\pm 5.52}$	$\underline{65.67}_{\pm 4.30}$	$\underline{40.92}_{\pm 4.04}$	$48.87_{\pm 5.05}$	$\underline{38.73}_{\pm 0.57}$	$18.96$
M3	$\underline{44.54}_{\pm 2.32}$	$\underline{53.89}_{\pm 0.86}$	$\underline{43.72}_{\pm 7.79}$	$65.44_{\pm 3.02}$	$40.57_{\pm 7.63}$	$\underline{48.97}_{\pm 3.17}$	$38.71_{\pm 0.43}$	$19.69$
M4	$43.46_{\pm 2.43}$	$52.44_{\pm 1.22}$	$42.39_{\pm 7.18}$	$64.93_{\pm 4.11}$	$40.49_{\pm 3.06}$	$47.96_{\pm 3.22}$	$38.12_{\pm 0.62}$	$16.30$
M5	$42.83_{\pm 2.08}$	$52.25_{\pm 1.61}$	$42.83_{\pm 5.19}$	$64.23_{\pm 4.53}$	$28.67_{\pm 1.29}$	$46.77_{\pm 3.38}$	$37.87_{\pm 0.66}$	$7.98$

Table 2: Core component ablations:

F1_{\mathrm{event}}

on eight EEG datasets (mean

\pm

SD over five cross-validation folds; TUSZ: a single result on its predefined split). Best and second-best scores are bold and underlined , respectively.

Configuration		P300		Motor Imagery		Motor Execution		Speech	Seizure
Quantization	FOCAL	BI2014b	BI2013a	BNCI2014	BNCI2015	BNCI2020	HDOP	Broderick	TUSZ
✗	✗	$42.02_{\pm 1.38}$	$50.64_{\pm 0.65}$	$39.44_{\pm 3.27}$	$63.05_{\pm 3.50}$	$36.73_{\pm 3.61}$	$46.34_{\pm 2.79}$	$37.15_{\pm 0.30}$	$12.24$
✓	✗	$43.50_{\pm 2.74}$	$51.11_{\pm 1.32}$	$41.23_{\pm 6.30}$	$64.15_{\pm 5.11}$	$37.28_{\pm 4.54}$	$46.50_{\pm 1.43}$	$\underline{37.98}_{\pm 0.44}$	$16.32$
✗	✓	$\underline{43.98}_{\pm 2.64}$	$\underline{51.99}_{\pm 1.33}$	$\underline{42.76}_{\pm 5.88}$	$\underline{65.34}_{\pm 2.88}$	$\underline{37.79}_{\pm 3.70}$	$\underline{47.98}_{\pm 3.04}$	$37.94_{\pm 0.51}$	$\underline{20.14}$
✓	✓	$\mathbf{45.23}_{\pm 1.22}$	$\mathbf{56.50}_{\pm 0.97}$	$\mathbf{45.55}_{\pm 4.96}$	$\mathbf{67.77}_{\pm 3.99}$	$\mathbf{41.85}_{\pm 3.83}$	$\mathbf{52.91}_{\pm 3.33}$	$\mathbf{38.76}_{\pm 0.57}$	$\mathbf{25.06}$

Table 3: Quantization and FOCAL ablation:

F1_{\mathrm{event}}

(%, five-fold mean

\pm

SD; TUSZ: a single result).

Panel of Figure 3 — Figure 3: Temporal decoding examples from (a) BNCI2020 and (b) TUSZ. Colored intervals indicate event classes and temporal extents; GT denotes ground truth. Side panels report metrics computed on the displayed segments: eIoU is the mean temporal IoU over overlapping, same-class one-to-one matches, and event-level F1 ( $F1_{\mathrm{event}}$ ) is evaluated at an IoU threshold of 0.5. N/A indicates that no matches are available for computing eIoU.

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 4: Event-level F1 of NeuroDETR and baseline methods at different temporal IoU thresholds. Higher thresholds impose stricter temporal localization requirements.

Figure 5: Event-level false negative rate (FNR) across eight datasets. Lower values indicate fewer missed events. NeuroDETR achieves the lowest FNR on six datasets. Error bars denote standard deviations and are omitted for TUSZ, where standard deviations are unavailable.

Figure 6: Event-level false discovery rate (FDR) across eight datasets. Lower values indicate a smaller proportion of incorrect detections among predictions. NeuroDETR achieves the lowest FDR on six datasets. Error bars denote standard deviations and are omitted for TUSZ, where standard deviations are unavailable.

Dataset	Decoder queries / Event F1
BI2014b	50	100	150	200	300
	44.00	45.13	45.23	44.20	43.20
BI2013a	50	100	150	200	300
	55.65	52.71	56.50	52.46	54.28
BNCI2014	1	3	5	10	20
	42.55	43.10	45.55	45.44	42.77

Appendix

Table 4: Sensitivity to decoder query count

N_{e}

. For each dataset, the upper and lower rows report

N_{e}

and

F1_{\mathrm{event}}

, respectively. Bold indicates the selected query count and its corresponding score.

Coarse region ( $S_{2}$ )	Fine clusters ( $S_{3}$ )	Count
Frontal	Left, midline, and right frontal	3
Temporal	Left and right temporal	2
Central	Left, midline, and right central	3
Parietal	Left, midline, and right parietal	3
Occipital	Undivided occipital group	1
Total		12

Appendix

Table 5: Coarse-to-fine spatial grouping.

Dataset	Domain	Subjects	Channels	Event (h)	Non-event (h)	Mean event duration (s)	Event coverage (%)	Split	Window (s)	Batch size per GPU
Korczowski2014b	P300	38	32	1.0	1.3	1.0	44	CV	8	32
Vaineau2018Brain	P300	24	16	4.4	1.1	1.0	80	CV	32	8
Tangermann2012	Motor imagery	9	22	5.8	5.8	4.0	50	CV	8	32
Faller2012	Motor imagery	12	13	7.8	8.9	5.0	47	CV	32	4
BNCI2020	Motor execution	45	12/32/58	10	27	5.0	27	CV	32	4
HDOP	Motor execution	14	62	0.24	3.7	0.20	6.1	CV	9	8

Appendix

Table 6: Dataset statistics and training configurations.

Module	Input shape	Output shape	Configuration
Spatial channel mixing	$B\times C\times T$	$B\times 270\times T$	Fourier embedding dimension: 2048; spatial dropout radius: 0.2
Pointwise projection	$B\times 270\times T$	$B\times 256\times T$	Conv1d, kernel size: 1
Input dropout	$B\times 256\times T$	$B\times 256\times T$	Dropout probability: 0.1
Temporal stage 1	$B\times 256\times T$	$B\times 512\times T$	Kernel size: 9; dilation: 1
Temporal stages 2–9	$B\times 512\times T$	$B\times 512\times T$	Kernel size: 9; dilation: $d_{\ell}=\lfloor 2.5^{\ell-1}\rfloor$
Temporal stage 10	$B\times 512\times T$	$B\times 128\times T$	Kernel size: 9; dilation: 3814; no normalization or activation

Appendix

Table 7: Backbone architecture and hyperparameters. The temporal resolution is preserved throughout the backbone.

Electroencephalography

Electroencephalography (EEG) studies brain-generated electrical activity to understand cognition and support diagnosis, monitoring, and brain-computer interfaces.

See all 183 papers

May 11, 2026

DANCE: Detect and Classify Events in EEG

Event identification in continuous neural recordings is a critical task in neuroscience. Decoding in EEG is dominated by classifying windows aligned to known event onsets. However, while available in controlled experiments, such onsets are absent in continuous real-world monitoring. Here, we introduce DANCE, a deep learning pipeline that frames neural decoding as a set-prediction problem and jointly detects and classifies events directly from raw, unaligned signals. Evaluated separately on ten datasets curated from the literature with a wide variety of event types (ranging from milliseconds to minutes in duration), our model outperforms existing methods on a broad range of cognitive, clinical and BCI tasks. This single architecture establishes a new state of the art in the competitive task of seizure monitoring and matches the accuracy of onset-informed models for BCI tasks. Overall, our method marks a step towards end-to-end asynchronous neural decoding models

Jarod Lévy, Hubert Banville, Jérémy Rapin +3

1Meta AI · 2Inria, Université Paris-Saclay, Palaiseau, France

Electroencephalography Brain-Computer Interfaces EEG Decoding Set Prediction Classification Structured Prediction

Ask PDF

Jun 18, 2026

Neural Events: Discrete Asynchronous Autoencoders for Event-Based Vision

Event cameras capture dynamic scenes with exceptional temporal fidelity by representing them as a continuous stream of microsecond resolution events. Each individual event, however, only carries minimal semantic value, merely signaling a localized brightness change. To derive meaningful signals, downstream algorithms need to quickly integrate cues from a potentially massive torrent of low-information events. Current architectures, however, are easily overwhelmed, struggling to balance capturing fine-grained temporal dynamics and maintaining a manageable data throughput. This paper proposes a framework to re-tokenize event streams into a small set of highly informative neural events, each representing a local spatio-temporal context window with a discrete learnable code. Every time this code flips, a neural event is triggered, yielding a highly compressed data stream. We demonstrate that, across object detection and classification, networks trained on neural events are on par or surpass the performance of state-of-the-art approaches while reducing the event rate by a factor of 2.0.

Roberto Pellerito, Daniel Gehrig, Shintaro Shiba +1

Robotics and Perception Group, University of Zurich · University of Pennsylvania · The University of Tokyo +1

Event-Based Vision Representation Learning Autoencoders Event-Based Object Detection

Ask PDF

Jun 9, 2026

Transformer Based Model for Spatiotemporal Feature Learning in EEG Emotion Recognition

Electroencephalography (EEG) is a widely adopted technique for monitoring brain activity, offering valuable insights into neurological states due to its high temporal resolution and cost-effectiveness. To enhance the analysis of complex EEG data, we propose EEG-TransNet, an architecture designed to capture temporal, regional, and synchronous features of EEG signals. EEG-TransNet introduces three key modules: 1) a preprocessing and feature extraction module leveraging ResNet and wavelet-based denoising, 2) a Local Self-Attention Block for regional feature learning, and 3) a Fuzzy-Attention Synchronous Transformer (FAST) to model spatiotemporal dependencies. Through extensive experiments on three EEG datasets (BETA, SEED, and DepEEG), the proposed model consistently outperforms other methods in terms of classification accuracy and robustness across varying signal lengths. Ablation studies confirm the contribution of the Local Self-Attention Block in improving performance, and the inclusion of depthwise separable convolutions in the decoder reduces computational complexity while maintaining high accuracy. EEG-TransNet's ability to generalize across subjects with minimal performance variation highlights its potential as a robust tool for EEG-based brain activity classification and emotion recognition tasks.

Xinglong Cui, Dian Gu

Beijing Neurodeep Technology Co., Ltd, Beijing, 100176, China · University of Pennsylvania, Seattle, 98121, USA

Electroencephalography Temporal Attention EEG Emotion Recognition Time Series Transformers

Ask PDF

Abstract

Electroencephalography

Explore similar work

DANCE: Detect and Classify Events in EEG

Neural Events: Discrete Asynchronous Autoencoders for Event-Based Vision

Transformer Based Model for Spatiotemporal Feature Learning in EEG Emotion Recognition